UDK

(This series of articles is about the capabilities of GViz, a machinima toolkit for UDK.)

The ingame shader for the characters is mostly done.
It can create generic materials:

(Yeah, Victor sometimes resembles another guy also on the ICAO alphabet…)

But it also has features specifically made for virtual actors:

The most apparent of those is the clothing: the different parts of the garment can be independently colored or removed.

To achieve this, the mesh has a second UV set (along with a full unwrap) which maps a small texture of 8 different colors:

Each color identifies a piece of clothing. The shader creates black and white masks from these colors using basic math then blends in the related color parameters of the current material instance.

The order of colors is important: the border between masking colors usually appear in the middle of polygons as opposed to polygon edges. This way the position and shape of the end of clothing pieces are not restricted by the topology of the mesh. You can see that on the zoomed in view of the knee: the shorts has an edge pretty different from the edge flow of the geometry.

The bilinear filtering is turned off so no blending occurs between the colors, as it screws up the mask generation later on.

The arrows show the order of clothing pieces: Black means uncovered skin, white is “cap” (or hair). Red is the T-shirt which has two neighbors, the green long sleeves and the cyan pants, and so on.

Otherwise the character uses 3 more, grayscale textures: diffuse map (only in the case of the camo guy), a cavity map and an accessibility map. The latter is a combination of 3DS Max’s subsurface scattering approximation and ambient occlusion maps.

Another (optional) texture is for the “logo”, some kind of a pattern on the clothes. It’s basically a distance field image baked onto the low poly geometry. The DF was generated from black and white patterns, in this case lines made in Illustrator.

Since it’s a distance field, changing the thickness and border smoothness of the shapes are easy. In theory all the fancy stuff I wrote about earlier could be implemented but I’m not sure it will be necessary.

Now I go back to fix a few things in the geometry then it’s time to make the final skeleton and the skinning.

(This series of articles is about the capabilities of GViz, a machinima toolkit for UDK.)

After last weeks cubemap reflections I’d like to show you one more thing captured cubemaps can be used for: image based lighting.

We can illuminate the scene with an image, an emissive texture which is in this case a cubemap. The CM was captured so it contains the look of the distant environment or skybox. Using that color information we can compliment the effect of direct lights and end up with more natural shading.

Here is an example scene with a static mesh and a dynamic skeletal mesh, using only a single DominantDirectionalLight:

Direct lighting only

The shadows are articulated but way too deep. The main light direction and color is okay but beyond that there is little visual relation to this environment:

Part of the cubemap showing the environment

This captured environment cubemap is projected onto an inside-out sphere. This “light sphere” encompasses the playable area but still located inside the LightmassImportanceVolume.

Using the light sphere only the scene looks like this:

Indirect (image) lighting only

The environment’s contribution is obvious, but the lighting is very diffuse without distinct shadows. The skeletal mesh’s shadow is not consistent with the static ones.

Now, let’s use both lighting methods:

Combined direct and image based lighting

We got proper shadows on both dynamic and static geometry and natural colors filling areas not directly lit by the distant light.

The contrast, brightness and saturation of the environment texture can be adjusted for maximum control over the final look.

Since it’s a relatively small sphere with a texture, it’s much faster to compute than including the skybox and all the related decoration meshes in the GI rendering.